On September 26, @DhravyaShah open-sourced what his team calls a "company brain": an AI employee that lives in Slack, holds the knowledge of the whole company and deploys in one click (post). The post passed 405,000 views and 1,784 engagements in a day.

If you run a Shopify store with one to five people, you probably use AI the scattered way: one chat for product descriptions, another for support replies, another for ad copy, each starting from zero. The idea spreading this month is to give every AI tool the same written knowledge about your business. This piece covers why that idea took off, what a store's version should contain, where people still matter, and what goes wrong when the shared knowledge is wrong.

One brain for every agent became the default idea this month

Three very different players moved in the same direction within a few weeks. In late August, HubSpot co-founder @dharmesh launched YouSpot, a CRM for one-person companies that builds a "second brain" from your business context, at an intro price of $1 a month for the first 1,000 customers (post). His announcement reached 1.3 million views.

In mid-September, HubSpot itself rebuilt its platform around what it calls Growth Context, which combines information about a company, its teams and its customers so its assistant can pick the right agents for a task (martech.org).

Smaller operators got there on their own. @coreyganim runs his business with a "Second Brain" stored in GitHub: every new agent gets read and write access to it, and a weekly skill pushes what the agent learned back into the shared files (post).

@Hey_karl put the reason plainly in a post about connected wellness agents: the goal is fewer tools that remember the same person, so the shopping agent does not start from zero after the meal planner has done its work (post).

What a store's brain should hold: six files to start

The most practical template comes from @shannholmberg, in a post with 2,062 engagements (post). He suggests a handful of plain text files, built from material you already have, such as sales calls, proposals and customer interviews.

Mapped to a Shopify store, the six files look like this:

  1. company: what you sell, who it is for, where you ship.
  2. customer: their problems, buying triggers, objections and the exact words they use in reviews and support emails.
  3. offer: products, prices, bundles, shipping and return terms you can actually honor.
  4. positioning: why someone buys from you instead of Amazon or a bigger brand.
  5. voice: how your brand writes, with examples and phrases to avoid.
  6. proof: reviews, test results and the claims you are allowed to make.

His template was written for marketing teams, not stores. Adding product specs, your current return policy and your top support questions to the offer and customer files is a sensible adaptation, but no one has yet published a before-and-after test of this setup on a real Shopify store.

@shannholmberg's follow-up adds two layers once the basics work: a data warehouse so agents can see what performed, and a brand book so their design choices follow your taste (post). An ecommerce-flavored example comes from @RoundtableSpace, which describes a knowledge base that stores winning ad structures, sales scripts and character sheets for a reusable ad agent (post).

A shared brain does not decide who you should hire

A shared brain removes repeated work. It does not settle whether you need people. @henrydaubrez, after 20 years in marketing, argues that "mediocrity is very often a group project," pointing at too many opinions and approval layers (post).

@sonalshukla3377 frames the problem more narrowly: most founders are not short of AI, they still do too many things by hand, like writing SOPs (standard operating procedures) and answering the same customer question again (post).

The other side has evidence too. @yamanzdh wrote that investors wanted his AI-native services startup to remove the expert from the loop, but six months of operating showed clients pay for the opposite (post). That post drew 131,000 views. @Zephyr_hg's one-person company plan keeps a person on every judgment call and automates only the parts that repeat (post).

Be careful with the viral numbers in between. @eng_khairallah1's story of seven Claude agents producing $18,800 a month on $480 of API costs is widely shared, but it comes with no verifiable source (post). There is also no public data comparing solo AI-run Shopify stores with staffed ones on growth or margin, and most cases cited are services or software businesses.

When the shared brain is wrong, every agent is wrong

Sharing knowledge also shares mistakes. @beamnxw described a setup where 64 agent sessions used one record of past decisions: across 178 checks over three days, 68% confirmed an earlier decision and about 4.5% changed what the agent did (post). In one case an agent removed a database index it judged unused, and a human had to re-test and reverse it.

That example comes from software, and there is no public data yet on how often store agents act on stale information. The store version is easy to picture, though: if your offer file still lists last season's return window, your support agent, your FAQ writer and your email agent will all repeat it.

The systems built for commerce handle this the same way. @sripathiteja4 describes Polar Operator reading ads, email and inventory data but requiring a human to approve any change in Meta, Klaviyo or Shopify, summed up as "Ask first, act second." (post).

Anthropic's merchant agent blueprint proposes price and inventory changes and waits for approval instead of applying them, as @oikon48 notes (post). Creatify promotes its media-buying agent on the same promise: every change drafted, none applied without a click (post).

Build your store's brain this week

  1. Day 1: Export your last 100 reviews and 50 support emails. Paste recurring phrases into the customer file.
  2. Day 2: Write the offer file from your live Shopify settings: prices, bundles, shipping zones, return window. Add the date of your last update at the top.
  3. Day 3: Write company, positioning and voice in one sitting. Keep each under one page.
  4. Day 4: Build the proof file only from claims you can back up.
  5. Day 5: Point every AI tool you use at the same folder. Regenerate one product description and one support reply, and compare them with the versions from before.
  6. After that, every week: Update the files whenever a price, policy or product changes, and keep a person approving anything that touches money, prices or stock.

Sources